How to Probe: Simple Yet Effective Techniques for Improving Post-hoc Explanations
This work addresses the reliability of model explanations for practitioners, revealing a critical oversight in post-hoc attribution methods that can affect interpretation across diverse pre-training techniques.
The paper challenges the assumption that post-hoc explanations are independent of training details, showing that the classification layer's training (less than 10% of parameters) strongly influences explanation quality, and proposes adjustments to improve it.
Post-hoc importance attribution methods are a popular tool for "explaining" Deep Neural Networks (DNNs) and are inherently based on the assumption that the explanations can be applied independently of how the models were trained. Contrarily, in this work we bring forward empirical evidence that challenges this very notion. Surprisingly, we discover a strong dependency on and demonstrate that the training details of a pre-trained model's classification layer (less than 10 percent of model parameters) play a crucial role, much more than the pre-training scheme itself. This is of high practical relevance: (1) as techniques for pre-training models are becoming increasingly diverse, understanding the interplay between these techniques and attribution methods is critical; (2) it sheds light on an important yet overlooked assumption of post-hoc attribution methods which can drastically impact model explanations and how they are interpreted eventually. With this finding we also present simple yet effective adjustments to the classification layers, that can significantly enhance the quality of model explanations. We validate our findings across several visual pre-training frameworks (fully-supervised, self-supervised, contrastive vision-language training) and analyse how they impact explanations for a wide range of attribution methods on a diverse set of evaluation metrics.